Pass a top-level JS null TurboModule arg to ObjC as nil, not NSNull (#58190)

[christophpurrer]

Summary:

Changelog: [iOS][Fixed] - Pass a top-level JS null TurboModule argument to Objective-C as nil instead of NSNull when enableModuleArgumentNSNullConversionIOS is enabled

When enableModuleArgumentNSNullConversionIOS is on, convertJSIValueToObjCObject
maps a JS null to (id)kCFNull. That is the intended behaviour for nulls nested
inside arrays and dictionaries, but a null in argument position must still reach
Objective-C as nil — NSNull is truthy and does not respond to the selectors the
receiver expects, so leaking it crashes the callee.

The guard that enforced this lived three branches deep in
ObjCTurboModule::setInvocationArg, reachable only when all of the following held:

• objCArgType == encode(id), and
• getArgumentTypeName(...) returned non-nil, and
• RCTConvert responded to a selector named after that type.

getArgumentTypeName resolves the argument type by scanning for __rct_export__-prefixed
selectors, which the compiler only emits for methods declared with RCT_EXPORT_METHOD.
Any TurboModule method without that macro — or with an id-typed argument, since
[RCTConvert respondsToSelector:selector(id:)] is NO — silently skipped the guard and
received NSNull.

This diff hoists the check to immediately after the conversion, so it applies to every
argument regardless of the method's __rct_export__ metadata, its ObjC type encoding, or
whether an RCTConvert converter exists. Returning without calling setArgument: leaves
the NSInvocation slot zeroed, i.e. nil — identical to what the old guard did.

Behaviour is unchanged when the flag is off: the check short-circuits on the flag. Nested
NSNull inside arrays and dictionaries is untouched, as asserted by the new
testInvokeTurboModuleKeepsNestedNullAsNSNullWhenFlagEnabled case. (The flag-enabled
branch of the pre-existing testInvokeTurboModuleWithNull case never executes while the
flag defaults to false, so it did not cover this.)

The pre-existing check inside the RCTConvert branch is left in place. It is now
effectively unreachable — objCArg == kCFNull is the only way convertedObjCArg can be
kCFNull, because every RCTConvert converter either returns a non-kCFNull input
unchanged or builds a new object — but it costs nothing and keeps the diff narrow.

Differential Revision: D117960577

[meta-codesync]

@christophpurrer has exported this pull request. If you are a Meta employee, you can view the originating Diff in D117960577.